Exploiting write semantics in implementing partially replicated causal objects

To support efficient access to objects that support cooperation between users, the objects need to be replicated at nodes where they are frequently accessed. Causal consistency is a consistency criterion particularly attractive for objects shared in cooperative applications. It is based on a causality relation on read and write operations issued by concurrent processes. It only requires that writes that depend on one another be perceived in their dependence order by all processes; so, two independent writes can be perceived in a distinct order by two different processes. In message-passing systems causal consistency is usually implemented by a protocol that ensures causal delivery of all messages. In this paper we develop a new protocol for causally consistent shared objects that exploits the write semantics that can be associated with update messages. This protocol allows early delivery of update messages and can discard messages carrying obsolete values. By not requiring that all update messages be delivered in causal order this protocol improves efficiency, reduces buffering requirements and compares very favorably with protocols that implement causal objects based on causal delivery of all update messages.